MONROE, Conn. — Changlin Li, executive director of the A.I. Safety Awareness Project, spoke on the topic of A.I. during an appearance at the Monroe Senior Center June 16, but before talking about potential problems and threats posed by the emerging technology, he demonstrated some of the benefits.
Li told center patrons to think of artificial intelligence (A.I.) as an overeager intern, before typing questions on ChatGPT. Then he used an app on his phone to speak with ChatGPT rather than type.
“Hi ChatGPT, can you hear me?” Li asked, while on speakerphone.
“Yes Changlin, I can hear you,” the bot responded in a friendly male voice. “What’s on your mind?”
Li said he was planning a trip to the Bahamas from Monroe. The Chat bot said it would check flights. “Do you want something relaxing? I can plan a snorkeling trip,” it said.
Li said he wanted to leave from Westchester Airport on July 31. The bot told him there would likely be a connecting flight in Miami and that he should reach his destination by Aug. 1. “I recommend some beach excursions,” it said. “Easy, breezy.”
“It’s much more like we interact with humans,” Li said. “This is where we can run into some problems.”
Criminals often use A.I. to mimic the voice of relatives and friends of their marks to trick them into sending money. During the demonstration, Li used a voice recording of this reporter, Monroe Sun Editor Bill Bittar, to show A.I. created audio clips, including an angry voicemail and a recording in which Bittar was a parent calling his child’s school to tell the main office to allow “a family friend” to pick up his daughter.
Li also had A.I. generate fake photos using two family pictures Bittar submitted for the demonstration. In one A.I. generated photo, Bittar is wearing an orange prison jumpsuit while sitting in a jail cell, in another he’s wearing camo and holding a hunting rifle.
Sometimes imperfections can be found in an A.I. generated photo, so Li said online criminals may generate a picture that is slightly out-of-focus to hide the fine details. He showed what resembled a black-and-white surveillance camera screenshot of Bittar wearing a hoodie and looking intimidating while standing near a woman in an elevator.
Li also used A.I. to make realistic looking New York Times articles showing The Monroe Sun editor involved in a scandal. These tactics have been used against politicians on both sides of the aisle during presidential elections.
In a targeted attack, Li also showed how A.I. could create an identity of someone, along with a backstory, and generate persuasive phishing emails to Bittar’s LinkedIn profile to steal his password.
On June 16, Monroe Police Det. Kyle Stevens participated in the presentation, along with Officer Danny Leon. Also present from the A.I. Safety Awareness Project were Co-Director Madeline McBride and facilitator Lauren Mecca.
The A.I. Safety Awareness Project is a 501c3 nonprofit organization whose mission is for everyone to have knowledge, opinions and a voice in the direction of AI as a technology.
“We want to be there for the first step,” Li said. “We don’t want to be there for the second step. That’s up to you. Talk to your family and friends. What are you most excited about with A.I.? What do you want the future to be?”
Though AI can do some great things, Li said his nonprofit is a safety organization so it focuses on risk. The A.I. Safety Awareness Project does presentations with police, at libraries, city and state agencies and community centers.
What is AI?
Just as an “overeager intern” can sometimes make mistakes, so does A.I., according to Li. When it does, A.I. is different from software, so a line of code cannot be checked for an error, he said. A.I. knowledge grows by learning from massive amounts of data. Li suggests fact-checking A.I. just as one would an intern, looking at different sources to verify information.
AI is trained, not designed, Li said. “Even their creators do not know how they produce their outputs,” he said. “It’s much more a process of growing something than engineering something. There is no universal official definition of A.I.”
To give a working definition, he defined it as machines mimicking aspects of human intelligence.
Li said the first form of A.I. was the thermostat in the 1940s, then in 1997 Deep Blue defeated the chess champ. But it was only programmed to play chess. In 2011 Watson won “Jeopardy!” and in 2016 AlphaGo defeated Go Champ. DALLE Image generator came in 2021, and ChatGPT had an explosion of capabilities in 2022.
Li said there is a Large Language Model (LLM) of A.I., a system trained on huge amounts of text to understand and generate human-like language by modeling the next word.
There are also Generative A.I. systems, which came on the scene around 2022. He said it creates new content (text, images, videos) from existing data and generates original outputs based on prompts.
“It can write a song for you,” Li said.
To illustrate how far A.I. has come in a short period of time, Li showed an A.I. video generation of actor Will Smith eating spaghetti. The video showed a good likeness of Smith’s face, but it appeared a little goofy as he sloppily shoveled spaghetti into his mouth with his hands.
“This is awful, right?” Li asked. “This is the best we could do in March of 2023.”
Then he showed a video generation of a police traffic stop made in August of 2025. The crisp video showed a police officer walking up alongside a car. “You were doing 90 in a 40 zone!” he barked at the driver.
“I’m sorry, could we just smooth things over?” the driver asked as he handed the officer a wad of cash.
The officer took the money and said, “get out of here before I …” the audience at the senior center laughed at the scene.
“This is a completely fake video of a road stop,” Li said, adding the driver, the police officer and the car do not exist. “That was completely A.I. generated,” he said. “In the space of 36 months, it went from laughably bad to, ‘man, that looks really real!'”
Li said the speed of A.I. is also rapidly accelerating, from doubling the completion of tasks every seven months to doing so every four to five months.
Problems A.I. could bring
Li shared three pillars of concerns over A.I.:
- Will it exacerbate preexisting problems, such as fraud, scams, the concentration of power and loneliness?
- Will it make positions, such as a travel agent, obsolete? What does that mean for labor? Should we transition to a world where things are replaced by A.I.? If we do, how should it look?
- A loss of control. People who build the systems don’t understand why it does what it does. If we don’t know why systems do what they do, how do you make sure it doesn’t “go rogue” or kill everyone?
Resources are another issue. Data centers use a lot of energy for A.I., as well as water to cool the systems.
During a question-and-answer period after Li’s presentation, Detective Stevens told the audience how A.I. is making it more difficult for police to investigate fraud cases. He encouraged residents to contact police whenever they have questions about possible instances of fraud.
Seniors expressed concerns over the growing prevalence of A.I. and Li encouraged them to get involved by joining the mailing list for Aisafetyawarenessproject.org, letting their Congress member know A.I. safety is an important issue to them, and studying on their own. He also shared other resources like BlueDot Impact.
“I’m optimistic and idealistic about this,” Li said. “You don’t need a billion dollars to speak to your legislator.”
All respectful comments with the commenter’s first and last name are welcome.
